Several organizations and researchers are working on ways to identify fake geographic and satellite imagery as the ability to easily manipulate images threatens both national and economic security. The proliferation of deepfakes throughout civil society also has the potential to create safety hazards and inhibit humanitarian efforts, such as response to natural disasters if organizations unknowingly rely on false images and data.
The National Geospatial-Intelligence Agency (NGA), the government’s clearinghouse for satellite imagery, is working with other Intelligence Community partners to vet the validity of satellite imagery and spot and remove deepfakes from its catalog. NGA’s Chief Information Officer said recently that increasing cybersecurity and data integrity are among the agency’s top priorities, and that it has partnered with the National Reconnaissance Office (NRO), among others to work on the problem of deepfakes and imagery manipulation. NRO is responsible for obtaining commercial satellite imagery for use by NGA analysts.
In an article in Breaking Defense, NGA CIO Mark Andress described some of those efforts, namely algorithm development and data standards development.
“One is a no kidding methodology — algorithms — to vet imagery products. And we do that obviously in partnership with NRO, who manages much of our space-based commercial data integration. So we have a fantastic technology sharing agreement with the NRO and some research labs out there that are building this kind of the screening, if you will, for sources of data.”
Another key area of collaborative work on cybersecurity and deep fakes is “an emphasis on data standards,” Andress said. “That’s a no-brainer in the cybersecurity community. We’re all about data standards, reporting, all that type of stuff in the imagery world.”
He explained that data standards aren’t just important for cybersecurity and ensuring the quality of imagery and data, they are “equally important” for allowing NGA to speed products to customers and build a new kind of hybrid information architecture that is able to ingest and integrate data from myriad sources outside the IC.
Separately, Defense One reported that researchers from the University of Washington published a new study on a machine-learning tool that could significantly increase capability in detecting and eliminating fake satellite images by analyzing color, edge quality, and texture characteristics. The article noted that their tool correctly identified 94 percent of the fake images, but it also mischaracterized several real ones as fake, achieving an overall reliability of 73 percent.